Back in 2023, the AI market was powered almost entirely by imagination. With big enough model parameters, a strong compute narrative, and “pure” concepts, investors dared to assign high valuations. As long as you slapped on an “AI” label, attention and liquidity were never in short supply. Back then, people were debating whether “AI can achieve something”—almost nobody asked whether “AI can make money.”

Looking back now, the market has already taught a lesson with real money.

By 2026, investors are clearly pickier: even if the model is stronger, if it can’t run real workstreams, cut costs, or improve efficiency, it’s hard to keep getting funded. The competitive focus is shifting from “parameter scale” to “inference efficiency, execution capability, and business commercialization closed-loop.”

That’s also why AI Agents have suddenly been pushed to the forefront.

Most traditional AI tools stay at the level of “answering questions,” while the key change with Agents is that they can understand goals, call tools, break down steps, and actually get the job done. Data analysis, code development, content production, trade execution, and even the repetitive processes in enterprise operations are gradually being taken over by Agents. It’s no longer just an auxiliary tool—it’s starting to have the attributes of an “executor.”

For Web3, the imagination space this change unlocks is even more direct.

In the past, a large number of AI+Crypto projects essentially just attached an AI label to existing narratives—without solving real pain points on-chain or building a sustainable economic model. The directions with real potential are those that make AI an active participant in on-chain economic activity, not a bystander:

🔹Agents directly manage on-chain assets, automatically executing trades, rebalancing, or yield aggregation according to strategies;

🔹Agents participate in content production and distribution, completing value settlement through on-chain mechanisms;

🔹Agents help users carry out complex operations across protocols and across chains, forming new relationships of digital labor and collaboration.

Look at these directions—the core is actually the same: business logic is far clearer than the “AI concept” itself.

Of course,淘汰 will also accelerate. Projects without technical barriers, without real users, and that rely only on narrative financing will find it increasingly difficult to earn the market’s patience. In the next stage, the real standard by which the market votes with its feet won’t be whose story is better—it will be who can keep AI generating revenue.

Over the past few years, the market has completed the transition from “can AI work” to “can AI change an industry.”

Now the question has changed—more direct, and more brutal:

Can AI reliably make money?